The Evolution of Software Engineering: AI's Impact and the Productivity Paradox
The world of software engineering is undergoing a profound transformation, with AI taking center stage as a game-changer. But is it truly revolutionizing productivity, or are we witnessing a different kind of evolution? The debate rages on, and it's time to dive into the heart of the matter.
AI's Rise: Managers and Managers
Cameron Etezadi, CTO at LaunchDarkly and a former VP of engineering at IBM, makes a bold claim: every IC (Individual Contributor) engineer is now a front-line manager. This shift is attributed to AI's ability to automate tasks, requiring engineers to take on managerial responsibilities. Project planning, cross-team coordination, and decision-making are now integral to their roles. But does this mean they're more productive?
Gartner's research suggests a shrinking future for software engineering teams, with 60% of organizations expected to adopt smaller teams by 2029. Aliyah Camacho, a principal analyst at Gartner, predicts the rise of 'tiny teams' with as few as two to three engineers. If the role of software engineers is indeed morphing into middle management, then AI's impact on productivity should be evident.
However, the relationship between AI and productivity is not as straightforward as it seems. Tech leaders are questioning the metrics being tracked to assess AI's impact.
Productivity Metrics: More Code, Faster, But What?
Daniel Wang, CTO at Citizen Health and a former director of engineering at Uber, raises a critical point: productivity isn't solely about code output. He emphasizes that software exists to solve problems, and if AI enables us to ship 10 times more code without improving customer outcomes, we're not truly productive. Wang advocates for a shift in focus from code quantity to decision quality and outcomes.
He highlights metrics like cycle time, rollback rate, escaped defects, and system reliability as essential indicators of productivity. Qualitative questions, such as whether the chosen solution is the right one and whether the code addresses customer problems, should also be prioritized. Yet, these are often overlooked in favor of tangible but irrelevant metrics like lines of code, PRs merged, or velocity points.
Ameya Kanitkar, founder and CTO of Larridin, agrees that these traditional metrics are not a true gauge of value. He argues that rewarding teams for code output and commit frequency can lead to a volume-chasing mindset, potentially steering them away from customer-centric outcomes. Kanitkar's experience with AI agents creating both output and exhaustion further supports this argument.
The Productivity Paradox: Feeling Productive vs. Being Productive
The paradox arises when engineers feel productive despite potential decreases in actual productivity. David Holz, founder of Midjourney, shares a similar sentiment, expressing feelings of productivity and exhaustion with the latest coding models. This disconnect is exacerbated by the lack of reliable ways to track AI's impact on outcomes.
Kanitkar suggests that the constant context-switching required to manage multiple AI agents is causing fatigue. As engineering teams shrink, this constant agent-babysitting could become the new norm, but its productivity implications remain uncertain.
The Way Forward: Balancing Outputs and Outcomes
The key to navigating this evolving landscape lies in finding a balance between outputs and outcomes. While AI can enhance productivity, it's crucial to focus on decision quality and customer-centric metrics. By doing so, companies can ensure that AI's impact is not just about shipping more code faster but about delivering real value to customers.
In conclusion, the evolution of software engineering through AI is complex. It challenges traditional roles, metrics, and productivity measures. As we embrace AI's potential, we must also be mindful of its limitations and strive for a balanced approach that prioritizes both outputs and outcomes.
